Professional Documents
Culture Documents
Scaling Site
Scaling Site
Scaling Site
Abstract
arXiv:1603.03191v1 [math.AG] 10 Mar 2016
In this paper we construct the scaling site S by implementing the extension of scalars on
the arithmetic site A , from the smallest Boolean semifield B to the tropical semifield Rmax + .
The obtained semiringed topos is the Grothendieck topos [0, ∞) o N , semi-direct product of
×
the Euclidean half-line and the monoid N× of positive integers acting by multiplication, en-
dowed with the structure sheaf of piecewise affine, convex functions with integral slopes. We
show that pointwise [0, ∞) o N× coincides with the adele class space of Q and that this latter
space inherits the geometric structure of a tropical curve. We restrict this construction to the
periodic orbit of the scaling flow associated to each prime p and obtain a quasi-tropical struc-
ture which turns this orbit into a variant C p = R∗+ / pZ of the classical Jacobi description C∗ /qZ
of an elliptic curve. On C p , we develop the theory of Cartier divisors, determine the struc-
ture of the quotient Div(C p )/P of the abelian group of divisors by the subgroup of principal
divisors, develop the theory of theta functions, and prove the Riemann-Roch formula which
involves real valued dimensions, as in the type II index theory. We show that one would have
been led to the same definition of S by analyzing the well known results on the localization
of zeros of analytic functions involving Newton polygons in the non-archimedean case and
the Jensen’s formula in the complex case.
Contents
1 Introduction 2
1
4 The structure sheaf O of the scaling site 16
4.1 The stalks of the structure sheaf O . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 The points of S over Rmax+ . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 The sheaf of fractions and Cartier divisors . . . . . . . . . . . . . . . . . . . . . . . . 19
4.4 The order at a point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
Appendices 35
1 Introduction
In this article we show that the adele class space1 of Q admits a natural structure of tropical curve.
We follow the strategy outlined in [2, 3] and investigate the algebraic geometric structure of the
Scaling Site2 S obtained from the arithmetic site A by extension of scalars from the Boolean
+ (cf. Figure 1). As a Grothendieck topos S is described
semifield B to the tropical semifield Rmax
as [0, ∞) o N : the topos of N -equivariant sheaves (of sets) on the half-line and our first result
× ×
(Theorem 3.9) states that the isomorphism classes of points of this topos form the basic sector of
the adele class space of Q
Theorem. The space of points of the topos [0, ∞) o N× is canonically isomorphic to A (Rmax
+ ) '
× ∗
Q \AQ /Ẑ .
This result provides the missing geometric structure on the adele class space since the topos
[0, ∞) o N× inherits, from its construction by extension of scalars, a natural sheaf O of regular
functions. We call Scaling Site the semi-ringed topos
S := [0, ∞) o N× , O
1 More specifically the sector Q× \AQ /Ẑ∗ corresponding to the trivial Grossencharacter.
2 These results have been announced in [4].
2
θ
*
� � ��� /ℤ
so obtained. The sections of the sheaf O are convex, piecewise affine functions with integral
slopes. In Appendix C we review the well known results on the localization of zeros of ana-
lytic functions showing in which sense the tropical half-line (0, ∞), endowed with the sheaf of
convex piecewise affine functions with integral slopes, provides a suitable structure for the lo-
calization (both in the archimedean and non-archimedean case) of zeros of analytic functions in
the punctured unit disk. The new component supplied with the scaling site is the action of N×
by multiplication on the tropical half-line [0, ∞). On analytic functions this action is given by the
transformation f ( z) 7→ f ( zn ) for n ∈ N× , i.e. the action of the degree n endomorphism z 7→ zn
on the punctured unit disk.
The structure sheaf O of S is a sheaf of semirings of “characteristic one" (i.e. of semirings in
which 1 + 1 = 1) and the naturalness of this structure is justified at the conceptual level (see
Appendix B) by two facts. First, the endomorphisms of any object admit a canonical structure
of semiring in any category with finite products and coproducts when the canonical morphisms
from coproducts to products are isomorphisms. Second, passing from rings to semirings only
adds one more object to the list of finite fields Fq , namely the Boolean semifield B, and only
one object to the list of fields whose multiplicative group is cyclic, i.e. the semifield Zmax whose
multiplicative group is infinite cyclic. Both B and Zmax are semirings of characteristic one.
In section 4.1 we describe the stalks of the structure sheaf O and show (Theorem 4.3) that the
points S (Rmax+ ) of the scaling site defined over R+
max coincide with the points A (Rmax ) of the
+
arithmetic site defined over the same semifield. As stated in [2, 3] a long term goal of this project
is to develop an adequate version of the Riemann-Roch theorem in characteristic 1, suitable to
transplant the pRH proof of Weil to the Riemann zeta function. In this paper we test this idea by
restricting our geometric structure to the periodic orbits of the scaling flow, i.e. to the points of
[0, ∞) o N× over the image of Spec Z (cf. Figure 1 and [3], §5.1). We find that for each prime p
the corresponding circle of length log p is endowed with a quasi-tropical structure which turns
this orbit into a variant C p = R∗+ / pZ of the classical Jacobi description C∗ /qZ of an elliptic curve.
The structure sheaf O p of C p is obtained by restriction of O to C p and its sections are periodic
3
functions f ( pλ ) = f (λ ), λ ∈ R∗+ , which are convex, piecewise affine and whose derivatives
take values in the group H p ⊂ R of rational numbers with denominators a power of p. When
suitably understood in conceptual terms using Cartier divisors, the notions of rational functions,
divisors, etc. on C p are all meaningful. The global rational functions form a semifield K(C p ) (of
characteristic one). A new feature of this construction is that the degree of a divisor can be any
real number. We introduce an invariant χ( D ) ∈ Z/( p − 1)Z for divisors D on C p and determine,
in Theorem 5.6, the precise structure of the quotient Div(C p )/P of the abelian group of divisors
by the subgroup of principal divisors
Theorem. The map (deg, χ) is an isomorphism of abelian groups
We develop, in analogy with the non-archimedean version established in [18], the theory of
theta functions on C p , starting with the following infinite sums as the analogues3 of the infinite
products defining classical theta functions
∞ ∞
m
∑ 0 ∨ ( p−m λ − 1 ) .
θ (λ ) := ∑ ( 0 ∨ ( 1 − p λ )) +
0 1
We define θ-functions Θh,µ for h ∈ H p , h > 0 and µ ∈ R∗+ . They are obtained by applying to the
basic theta function θ (λ ) defined above the symmetries associated to the various incarnations
(arithmetic, relative, absolute, geometric) of the “Frobenius" operator in this context. This part is
discussed in details in §5.2. The main output (cf. Theorem 5.12) is provided by the following
Theorem. Any function f ∈ K(C p ) is canonically expressed in terms of theta functions associated to the
principal divisor of f , and a constant c ∈ R
For each divisor D on C p we define the corresponding Riemann-Roch problem with solution
space H 0 ( D ) := H 0 (C p , O( D )). We introduce the continuous dimension DimR ( H 0 ( D )) of this
Rmax -module using a limit of normalized topological dimensions and find that DimR ( H 0 ( D )) is
a real number. The topological dimension used in this part is the Lebesgue covering dimension
which assigns to any topological space X an integer dimtop ( X ) ∈ {0, . . . , ∞} counting the min-
imal overlap of refinements of open covers. The appearance of arbitrary positive real numbers
as continuous dimensions of DimR ( H 0 ( D )) is due to the density in R of the subgroup H p ⊂ Q
and the fact that continuous dimensions are defined as limits
n
DimR ( H 0 ( D )) := lim p−n dimtop ( H 0 ( D ) p )
n→∞
n n
of normalized dimensions p−n dimtop ( H 0 ( D ) p ) where H 0 ( D ) p is a natural filtration of H 0 ( D )
involving the p-adic norms of the derivatives. We interpret this result as the characteristic 1
counterpart of the statement for matroid C ∗ -algebras and the type II normalized traces as in [5].
3 We use the notation x ∨ y for the max of two real numbers
4
The continuous dimensions which affect arbitrary positive real values appear when passing to
the von Neumann algebra of type II obtained as the weak closure of the C ∗ -algebra using the
trace to perform the completion. Finally, in Theorem 5.17 we prove that the Riemann-Roch
formula holds for C p
Theorem. Let D ∈ Div(C p ) be a divisor, then the limit DimR ( H 0 ( D )) exists and one has the Riemann-
Roch formula:
By comparing the periodic orbit C p with a tropical elliptic curve and our Riemann-Roch theorem
with the tropical Riemann-Roch theorem of [1,8,14] and its variants we find several fundamental
differences. First, for an elliptic tropical curve C given by a circle of length L, the structure of the
group Div(C )/P of divisor classes is inserted into an exact sequence of the form 0 → R/ LZ →
deg
Div(C )/P → Z → 0 (cf. [14]), while for the periodic orbit C p the group of divisor classes
is Div(C p )/P ' R × (Z/( p − 1)Z). The second fundamental difference is the appearance of
continuous dimensions in our Riemann-Roch theorem. The source for these differences is seen
when one compares the structure sheaf of C p with that of the elliptic tropical curve C := R/ LZ,
L = log p. Let us use for C p the variable u = log λ, so that the periodicity condition f ( px) = f ( x)
becomes translation invariance by log p. Then the local sections of the structure sheaf of C p are
in particular piecewise affine in the parameter λ and this condition is expressed, in the variable
u, by the piecewise vanishing of ∆2 f , where ∆2 is the elliptic translation invariant operator
2
2 ∂ ∂
∆2 = λ , ∆2 ( f ) := ( Du2 − Du ) f , Du := . (1)
∂λ ∂u
On the other hand, the local sections of the structure sheaf of C are in particular piecewise affine
in the parameter u, and this condition is expressed, in the variable u, by the piecewise vanishing
of Du2 f . Thus one readily sees that the difference between the two sheaves is due to the presence
of the sub-principal term − Du in (1).
Notations
For any abelian ordered group H we denote by Hmax = H ∪ {−∞} the semifield obtained
from H by applying the max-plus construction i.e. the addition is given by the max and the
multiplication by the addition in H. Since Rmax is isomorphic to Rmax + by the exponential map
(cf. [9]) we shall pass freely from the “additive" notation Rmax to the “multiplicative" one Rmax
+ .
5
4
3
7
2 6
1
5
4
-4 -3 -2 -1 1
3
-1
-2
2 4 6 8 10
-3
Pλ,u := {( x, y) ∈ R2 | λx + y ≤ u}, Pv := {( x, y) ∈ R2 | x ≤ v}
`C (λ ) := inf{u ∈ R | C ⊂ Pλ,u }
and this function is given in terms of the finitely many vertices ( x j , y j ) of the polygon C by the
formula
`C (λ ) = max λx j + y j . (2)
j
Note that `C (λ ) is meaningful also for λ = 0 and that limλ →∞ `C (λ )/λ = max x j = inf{v ∈ R |
C ⊂ Pv }. One then obtains the required identification
Lemma 2.1. The map L : C 7→ `C where `C (λ ), for ∀λ ∈ R+ has been defined in (2), is an isomorphism
of Zmax ⊗ˆ B Rmax with the semiring R(Z) of continuous4 convex piecewise affine functions on R+ with
slopes in Z ⊂ R and only finitely many discontinuities of the derivative. These functions are endowed
with the pointwise operations of functions with values in Rmax .
4 Continuity is automatic for convex functions on open intervals, see [17], Theorem 3.2.
6
Proof. It follows by the formula (2) that the function `C belongs to R(Z) since the slopes x j ∈ Z,
and the discontinuities of the derivative only occur when there exists a pair of vertices ( xi , yi ) 6=
( x j , y j ) such that λ ( xi − x j ) = y j − yi . Moreover `C is convex by construction as a supremum
of finitely many affine functions. The map C 7→ `C is evidently injective. The surjectivity is
a consequence of the fact that an element f ∈ R(Z) is a finite supremum of affine functions
λ 7→ aλ + b with a ∈ Z and b ∈ R, and thus of the form `C where C is the convex hull of the union
of quadrants ( a, b) − Q. The pointwise operations of functions with values in Rmax are given for
the addition by the rule ( f , g) 7→ f ∨ g, where ( f ∨ g)(λ ) := f (λ ) ∨ g(λ ) = max{ f (λ ), g(λ )}.
This corresponds to the max in (2) and thus to the convex hull of the union in terms of the
convex sets C. This shows that the map L : C 7→ `C is additive. It is also multiplicative, i.e.
`C+C0 = `C + `C0 . This follows using (2) and the identity max( A + B) = max( A) + max( B)
holding for any two finite subsets A, B ⊂ R.
Proposition 2.2. (i ) Under the isomorphism L the endomorphism Frn ⊗ Id of Zmax ⊗ ˆ B Rmax corresponds
×
to the action of N on R+ by multiplication.
(ii) The following map identifies the half line [0, ∞) with the space of characters of R(Z)
Proof. (i ) We use multiplicative notations both for Zmax and Rmax and represent elements of
Zmax ⊗ˆ B Rmax in terms of finite sums ∑ q x j ⊗B q y j with x j ∈ Z and y j ∈ R. In these terms the
isomorphism L : Zmax ⊗ ˆ B Rmax → R(Z) is such that
(ii) Let ι ∈ R(Z) be the function ι(λ ) = λ. An element ρ ∈ HomRmax (R(Z), Rmax ) is uniquely
specified by ρ(ι) ∈ Rmax and since 0 ∨ ι = ι and ρ(0) = 0 (as the morphism ρ preserves the
multiplicative unit) one has ρ(ι) = λ ∈ [0, ∞) ⊂ Rmax . By multiplicativity one gets ρ(kι) = kλ
for any k ∈ Z and by Rmax -linearity that ρ(kι + y) = kλ + y for y ∈ R. By additivity one then
gets that for any f = ∨( x jι + y j ) ∈ R(Z), ρ( f ) = ρ(∨( x jι + y j )) = ∨(λx j + y j ) = f (λ ).
Remark 2.3. One has in general, for a semiring R of characteristic 1, a natural isomorphism
7
Ω ⊂ [0, ∞) including those of the form [0, a) for a > 0. The morphisms between two objects of
C are defined by
HomC (Ω, Ω0 ) = {n ∈ N× | nΩ ⊂ Ω0 }
if Ω 6= ∅. By definition HomC (∅, Ω0 ) := {∗} i.e. the one point set, for any object Ω0 of C . Thus
the empty set is the initial object of C . The following lemma shows that pullbacks exist in the
category C
nj
Lemma 2.4. Let Ω j 6= ∅ ( j = 1, 2) and consider two morphisms φ j : Ω j → Ω given by integers
n j ∈ HomC (Ω j , Ω). Let n = lcm(n j ) be their lowest common multiple, n = a j n j and let Ω0 :=
{λ ∈ [0, ∞) | a j λ ∈ Ω j , j = 1, 2}. Then Ω0 is an object of C , and if it is non-empty one has
a j ∈ HomC (Ω0 , Ω j ) and (Ω0 , a j ) is the pullback of the φ j . When Ω0 = ∅ the pullback of the φ j is the
initial object of C .
Proof. By construction Ω0 = ∩2j=1 a− 1
j Ω j is an intersection of bounded open intervals and is thus
kj
an object of C . Let ψ j : W → Ω j be morphisms such that φ1 ◦ ψ1 = φ2 ◦ ψ2 . If W 6= ∅ it
contains a λ 6= 0 and one has n1 k1 = n2 k2 = mn for a unique m ∈ N× . One has k j = ma j and
kj m
k j W ⊂ Ω j so that mW ⊂ Ω0 . Moreover the map ψ j : W → Ω j is the composite of W → Ω0 with
a j ∈ HomC (Ω0 , Ω j ). This shows that (Ω0 , a j ) is the pullback of the φ j . It also shows that if there
kj
exists W 6= ∅ and morphisms ψ j : W → Ω j such that φ1 ◦ ψ1 = φ2 ◦ ψ2 , then Ω0 6= ∅. Otherwise
i.e. if this implies W = ∅ then one easily sees that the empty set is indeed the pullback.
When the small category C admits pullbacks one can associate a Grothendieck topology J to a
basis K, i.e. a function which assigns to any object Ω a collection K (Ω) of families of morphisms
with codomain Ω by the condition
S ∈ J (Ω) ⇐⇒ ∃ R ∈ K (Ω), R ⊂ S.
The above three conditions on J are derived from the following three conditions on K: (cf. [13]
Definition III.2.2):
1. For any isomorphism f with range Ω the singleton { f } is a covering.
8
3. Given a covering (Ω j ) j∈ I of Ω and for each j ∈ I a covering Ωi j of Ω j , the composite Ωi j
is a covering of Ω.
Proposition 2.5. (i ) For each object Ω of C , let K (Ω) be the collection of all ordinary covers {Ωi ⊂
Ω, i ∈ I | ∪Ωi = Ω} of Ω. Then K defines a basis for a Grothendieck topology J on C .
(ii) The Grothendieck topology J is subcanonical.
(iii) The category Sh(C , J ) of sheaves of sets on (C , J ) is canonically isomorphic to the category of N× -
equivariant sheaves of sets on [0, ∞).
Proof. (i ) The only isomorphisms in C are the identity maps, thus one verifies 1. To check the
n
condition 2., we let Ω0 → Ω be a morphism in C and {Ωi ⊂ Ω, i ∈ I | ∪Ωi = Ω} a covering of
Ω. Then it follows from Lemma 2.4 that the pullback of the cover is given by
π 2 : Ωi ×Ω Ω 0 → Ω 0 ' n −1 Ωi ∩ Ω 0 → Ω 0 .
This defines a covering of Ω0 . Finally the condition 3. is a standard fact on ordinary covers of a
topological space (here chosen to be [0, ∞)).
(ii) We prove that any representable presheaf is a sheaf on (C , J ). For a fixed object Ω of C and
an arbitrary open subset U of [0, ∞) one sets
This determines the subsheaf of the constant sheaf N× which is given by the local condition
around λ: {n ∈ N× | nλ ∈ Ω}.
(iii) An N× -equivariant sheaf (of sets) on [0, ∞) gives by restriction to C an object of Sh(C , J ).
Conversely let F be an object of Sh(C , J ), and U an arbitrary open subset of [0, ∞). Let {Ωi ⊂
U, i ∈ I | ∪Ωi = U } be a covering of U by bounded open intervals. Take the limit lim F (Ωi ∩
←−
Ω j ) in Sets of the diagram F (Ωi ∩ Ω j ) indexed by pairs (i, j) ∈ I 2 , Ωi ∩ Ω j 6= ∅ and arrows
(i, j) → (i, i), (i, j) → ( j, j). This is the equalizer of the two maps
p`
∏ F ( Ωi ) ⇒ ∏ F ( Ωi ∩ Ω j ) .
i∈ I i, j
Since F is an object of Sh(C , J ) the above equalizer does not depend upon the choice of the
covering of U by bounded open intervals, and defines a sheaf of sets on [0, ∞) endowed with an
action of N× compatible with the action of N× on [0, ∞).
Remark 2.6. The setting HomC (∅, Ω) = {∗} is imposed if one wants the Grothendieck topology
J to be subcanonical. Indeed, any sheaf evaluated on the empty set gives the one point set {∗}.
9
3.1 Flatness and continuity
It follows from [13] Corollary VII.5.4, that the points of the topos Sh(C , J ) correspond by means
of an equivalence of categories to continuous, flat functors F : C −→ Sets. Moreover again
from [13] Theorem VII.6.3, one knows that a functor F : C −→ Sets is flat iff it is filtering, i.e. F
fulfills the three conditions reported in the following definition
Definition 3.1. A functor F : C −→ Sets is filtering iff
3. Given two morphisms u, v : C → D in C and a ∈ F (C ) such that F (u) a = F (v) a, there exists an
object B of C , an element b ∈ F ( B) and a morphism w : B → C of C such that u ◦ w = v ◦ w and
F (w)b = a.
By [13], Lemma VII.5.3, a flat functor F : C −→ Sets is continuous iff it sends covering sieves to
epimorphic families.
Lemma 3.2. Let F : C −→ Sets be a flat functor and let J be the Grothendieck topology on C generated
by a basis K. Then F is continuous iff for any object U of C and a covering (U j → U ) j∈ I ∈ K (U ) the
family of maps F (U j → U ) : F (U j ) → F (U ) is jointly surjective.
Proof. Since any covering sieve S ∈ J (U ) contains a covering from the basis, the condition of the
Lemma implies that F sends covering sieves to epimorphic families. Conversely let R ∈ K (U ),
then the associated covering sieve ( R) ∈ J (U )
( R) := { f ◦ g | f ∈ R, Dom f = Codom g}
10
h j ∈ V j ∩ H+ we let h ∈ H+ , n j ∈ N× be such that h j = n j h. Let V be an open interval containing
h and such that n j V ⊂ V j . This defines an object V of C . One has h ∈ V ∩ H+ = FH (V ) and
nj n
the morphisms u j : V → V j fulfill FH (u j )h = h j . This gives condition 2. Finally, if u : C → D,
m
v : C → D are two morphisms in C and a ∈ FH (C ) is such that FH (u) a = FH (v) a, then since
a > 0 one gets that n = m and thus the condition 3 holds.
(ii) Let FH (V ) := V ∩ H+ be the continuous flat functor associated to H ⊂ R. Given a point
λ ∈ (0, ∞), we let V j be a basis of neighborhoods of λ of bounded open intervals. Then one has
lim FH (U ) 6= ∅ ⇐⇒ ∩ FH (V j ) 6= ∅ ⇐⇒ λ ∈ H.
←−
U 3x
This shows that one can recover the subgroup H ⊂ R from the continuous flat functor FH .
Moreover it shows that a morphism of functors from FH to FH 0 exists iff H ⊂ H 0 and hence that
the isomorphism class of the point p H uniquely determines H ⊂ R.
[0, ∞) o N× . To this end we shall need to state first several technical lemmas.
Lemma 3.4. Let F : C −→ Sets be a continuous flat functor. Then the following facts hold
(i) Let U j ⊂ V ( j = 1, 2) be objects of C with U1 ∩ U2 = ∅. Then the images F (U j → V ) F (U j ) ⊂ F (V )
are disjoint.
(ii) Let V be an object of C and x ∈ F (V ). Then there exist a unique λ ∈ V such that for any object U of
C , U ⊂ V, containing λ one has x ∈ F (U → V ) F (U ) ⊂ F (V ).
Proof. (i ) Since F is flat it commutes with fibered products. The fibered product of the two maps
U j → V is the empty set. The object ∅ of C admits the empty cover. Thus by continuity one
has F (∅) = ∅. One can also give the following argument. Let a j ∈ U j with F (U1 → V )( a1 ) =
F (U2 → V )( a2 ) = z. Then the flatness of F gives an object W of C an element c ∈ F (W ) and
nj nj
morphisms W → U j such that F (W → U j )(c) = a j . By composition with U j → V one gets
nj
F (W → V )(c) = z. The third filtering condition on F implies, since N× is simplifiable, that
n1 = n2 which contradicts n j W ⊂ U j and U1 ∩ U2 = ∅.
(ii) We show the existence of λ, its uniqueness then follows from (i) since distinct points have
disjoint neighborhoods given by objects of C . Let V = ∪W j where W j is an increasing family of
bounded open intervals such that W j ⊂ W j+1 . Then the continuity of F gives an interval W = W j ,
with W ⊂ V such that x ∈ F (W → V ) F (W ), i.e. x = F (W → V ) z, z ∈ F (W ). Using a cover U
of W by bounded open intervals one obtains, by continuity of F, an interval I1 ⊂ W of diameter
< 1/2 and z1 ∈ F ( I1 ) such that z = F ( I1 → W ) z1 . By induction one gets a decreasing sequence
of intervals Ik ⊂ Ik−1 ⊂ W of diameter < 1/2k and zk ∈ F ( Ik ) such that zk−1 = F ( Ik → Ik−1 ) zk .
Let λ be the unique limit point of the sequence Ik , i.e. the limit of any sequence λk ∈ Ik . One has
λ ∈ W ⊂ V. Let U ⊂ V be an object of C containing λ. Then U is an open neighborhood of λ
and there exists k such that Ik ⊂ U. One then gets
x = F (W → V ) z = F ( Ik → V ) zk = F (U → V ) F ( Ik → U ) zk ∈ F (U → V ) F (U )
11
In what follows we shall denote by λV : F (V ) → V the map defined in Lemma 3.4.
Lemma 3.5. Let F : C −→ Sets be a continuous flat functor.
(i) The maps λV : F (V ) → V define a natural transformation of F with the functor C −→ Sets,
V 7→ V.
(ii) The maps λV : F (V ) → V are injective when 0 ∈
/ V.
n n
Proof. (i ) Let U → V be a morphism in C . Let x ∈ F (U ), y = F (U → V ) x ∈ F (V ). For any
object W ⊂ U of C containing λU ( x) one has x ∈ F (W → U ) F (W ) and thus
n n n
y ∈ F (U → V ) F (W → U ) F (W ) = F (W → V ) F (W ) = F (nW → V ) F (W → nW ) F (W ).
This shows that y ∈ F (nW → V ) F (nW ). Thus if λV ( y) 6= nλU ( x) one obtains a contradiction by
Lemma 3.4 (i ). We have thus shown that the following diagram commutes
n
F (U → V )
F (U ) / F (V )
λU λV
n /
U V
∃V | λ ∈ λV ( F (V )) ⇐⇒ ∀V 3 λ, λ ∈ λV ( F (V ))
(ii) The subset of (0, ∞) defined by the above condition is of the form H ∩ (0, ∞), where H ⊂ R is a rank
one subgroup.
Proof. (i ) It is enough to show the implication ⇒. Let V and x ∈ F (V ) such that λ = λV ( x). Then
for any object U of C , U ⊂ V, containing λ one has x ∈ F (U → V ) F (U ) ⊂ F (V ). Let then W be
an object of C containing λ. Let U = W ∩ V and z ∈ F (U ) such that x = F (U → V ) z. One has
by Lemma 3.5 (i ), λW ( F (U → W ) z) = λU ( z) and λU ( z) = λV ( F (U → V ) z) = λV ( x) = λ. Thus
one gets λ ∈ λW ( F (W )) as required.
(ii) Let E = {λ > 0 | λ ∈ λV ( F (V )) , ∀V 3 λ }. Let λ j ∈ E, j = 1, 2, and V containing the λ j , and
x j ∈ F (V ) such that λ j = λV ( x j ). By the flatness of F there exists an object W of C , an element
nj nj
c ∈ F (W ) and morphisms W → V such that F (W → V )(c) = x j . By Lemma 3.5 (i ), one has
λV ( x j ) = n j λW (c). By (i ) one has λW (c) ∈ E. This shows that given any two elements λ j ∈ E
there exists λ ∈ E and integers n j ∈ N× such that n j λ = λ j , j = 1, 2. Moreover Lemma 3.5 (i )
12
shows that λ ∈ E ⇒ nλ ∈ E, ∀n ∈ N× . Thus E is an increasing union of subsets of the form
λk N× . Let H = ∪λk Z be the corresponding increasing union of subgroups of R. Then H is a rank
one subgroup of R and E = (0, ∞) ∩ H by construction.
Lemma 3.7. Let F : C −→ Sets be a continuous flat functor.
(i) One has
∃V | 0 ∈ λV ( F (V )) ⇐⇒ ∀V, λV ( F (V )) = V ∩ {0}
(ii) If the above equivalent conditions do not hold then there exists a rank one subgroup H ⊂ R and an
isomorphism of functors F ' FH .
Proof. (i ) Let V be such that 0 ∈ λV ( F (V )). One then has 0 ∈ V. Moreover the proof of Lemma
3.6 (i ) applies and shows that for any object W of C containing 0 one has 0 ∈ λW ( F (W )). Let W
be an object of C and assume that some λ > 0 belongs to λW ( F (W )). Let U be an object of C con-
taining both V and W. Then by Lemma 3.5 (i ), one has {0, λ } ⊂ λU ( F (U )). Let x j ∈ F (U ) with
λU ( x1 ) = 0, λU ( x2 ) = λ. Then the flatness of F gives an object U 0 of C , an element c ∈ F (U 0 ) and
nj nj
morphisms U 0 → U such that F (U 0 → U )(c) = x j . By Lemma 3.5 (i ), one has λU ( x j ) = n j λU 0 (c).
But λU ( x1 ) = 0 implies that λU 0 (c) = 0 and this contradicts λU ( x2 ) = λ 6= 0. Thus it follows that
for any object W of C , λW ( F (W )) contains at most 0 and it does if and only if one has 0 ∈ W.
(ii) If the condition of (i) does not hold, it follows that for any object V of C one has 0 ∈ /
λV ( F (V )). It follows that the canonical map ρV = F (V ∩ (0, ∞) → V ) : F (V ∩ (0, ∞)) → F (V )
is surjective. But by Lemma 3.5, (ii ), and the commutation with the localization λV this map is
injective. By Lemma 3.6 (ii ) there exists a rank one subgroup H ⊂ R such that for any U not
containing 0 one has λU ( F (U )) = U ∩ H. By Lemma 3.5, (ii ), the composite
−1
λV ∩(0,∞) ◦ ρV : F (V ) → V ∩ (0, ∞) ∩ H
X (n)
X / X (4)
pU pV
n
F (U → V )
F (U ) / F (V )
(iii) With the above notations, the action of N× on X defines a point of the topos N
c× .
(iv) Let H be an abstract rank one ordered group. The following defines a flat continuous functor FH0 :
C −→ Sets
0
FH / V, FH0 (V ) = H+ if 0 ∈ V.
(V ) = ∅ if 0 ∈
13
Proof. (i ) If 0 ∈
/ V one has λV ( F (V )) = ∅ and hence F (V ) = ∅. Let U ⊂ V with 0 ∈ U. The
map F (U → V ) : F (U ) → F (V ) is surjective since λV ( F (V )) = {0}. Let us show that it is
injective. Let x j ∈ F (U ) be such that F (U → V )( x j ) = z. By the flatness of F there exists an
nj nj
object W of C , an element c ∈ F (W ) and morphisms W → U such that F (W → U )(c) = x j . One
nj
then has F (W → V )(c) = z and the flatness of F shows, since N× is simplifiable, that n1 = n2
so that x1 = x2 . Thus the map F (U → V ) : F (U ) → F (V ) is bijective and the projective limit
X := limW 30 F (W ) is such that all maps pV : X → F (V ) are bijective.
←− n
(ii) The inclusions U ⊂ V, nU ⊂ nV form a commutative square with the maps U → nU,
n
V → nV. It follows that the map from X to X such that
n
F (U → nU ) pU ( x) = pnU ( X (n) x)
is independent of the choice of U containing 0 and turns (4) into commutative diagrams.
(iii) The set X is non-empty since otherwise one would have F (V ) = ∅, ∀V. Let x j ∈ X and
u j ∈ F (U ) (0 ∈ U) such that pU ( x j ) = u j . By the flatness of F there exists an object W of C , an
nj nj
element c ∈ F (W ) and morphisms W → U such that F (W → U )(c) = u j . One has 0 ∈ W. Let
x ∈ X such that pW ( x) = c, using (4) one gets X (n j ) x = x j . Thus the action of N× on X verifies
the second filtering condition. We now check the third filtering condition. Let x ∈ X and n1 6= n2
be such that X (n1 ) x = X (n2 ) x. Let 0 ∈ U, u = pU ( x). Let V ⊃ n j U. One then gets an equality
of the form
n n
F (U →1 V )(u) = F (U →2 V )(u)
and the flatness of F shows that n1 = n2 . This shows that the action of N× on X verifies the three
filtering conditions.
(iv) One checks that FH0 is continuous, its flatness follows from the rank one property of H.
We are now ready to define a map Θ from points of A (Rmax+ ) to points of the topos [0, ∞ ) o N .
×
As shown in [3] Theorem 3.8, the points of the arithmetic site A over R+ form the union of
max
abstract rank one ordered group H we assign the point q H of [0, ∞) o N× associated to the flat
continuous functor FH0 of Lemma 3.8, (iv ). Next to the point of A (Rmax ) \ A (B) corresponding
+
to the rank one subgroup H ⊂ R, we associate the point p H of Proposition 3.3.
Theorem 3.9. The map Θ defines a canonical isomorphism of A (Rmax
+ ) with the points of the topos
[0, ∞) o N .
×
Proof. The map Θ is well defined. Lemma 3.8 shows that is bijective from points of A (B) ⊂
A (Rmax
+ ) to points of [0, ∞ ) o N such that the associated flat continuous functor F : C −→ Sets
×
fulfills the hypothesis of Lemma 3.8. Proposition 3.3 shows that Θ is injective from points of
A (Rmax
+ ) \ A (B) to points of [0, ∞ ) o N . Finally, Lemma 3.7 shows that Θ is surjective.
×
14
3.4 Representation of points as filtering colimits of representables
Let C be a small category. Any covariant functor F : C −→ Sets which is representable, i.e. of
the form y I (C ) = HomC ( I, C ) for some object I of C is flat, and any flat functor is obtained as
a filtering colimit of such representable functors. In this subsection we describe such represen-
tations for the flat continuous functors associated to points of the topos [0, ∞) o N× . This result
will be used in the description of the stalks of the sheaves given in § 4.
Lemma 3.10. Let H ⊂ R be a rank one subgroup of R. Let hi ∈ H+ be a sequence of elements, n j ∈ N× ,
such that n j h j+1 = h j and that H = ∪h j Z. Let I j be bounded open intervals with h j ∈ I j for j ≥ 1, such
that
k−1
n j I j+1 ⊂ I j , ∀ j ≥ 1, ( ∏ ni ) Diameter( Ik ) → 0, when k → ∞. (5)
1
Then the limit lim y I j of the representable functors y I j (V ) := HomC ( I j , V ) defines the point p H of the
−→
topos [0, ∞) o N× .
Proof. We show that the functor lim y I j is simply given by
−→
lim y I j (V ) = lim(h− 1
V ∩ N× ) ∼ lim(V ∩ h j N× ) = V ∩ H+ .
−→ −→ j −→
It is enough to prove that the natural inclusion, due to h j ∈ I j
lim y I j (V ) ⊂ lim(h− 1
V ∩ N× )
−→ −→ j
(in both cases one uses multiplication by n j to organize the inductive system) is a bijection.
j−1
Indeed, let n ∈ h− 1 ×
k V ∩ N : we show that for j large enough one has n ∏k ni ∈ y I j (V ). Let
> 0 be a positive real number such that the neighborhood W of hk n of radius is contained
j−1
in V. Then for j large enough, using the hypothesis (5), one gets that n ∏k ni I j ⊂ W (since it
j−1
contains hk n and is of small enough diameter) and hence n ∏k n i ∈ y I j (V ).
Thus f fulfills the condition (iii ) of [13] Lemma VII.5.3, and f ∗ is the pullback part of the point
p H of the topos [0, ∞) o N× .
15
Next, we consider the limit case of Lemma 3.10 when all the h j are 0.
Lemma 3.12. Let n j ∈ N× , and I j be bounded open intervals containing 0 ∈ [0, ∞), for j ≥ 1, such that
(5) holds. Then the limit lim y I j of the representable functors y I j (V ) := HomC ( I j , V ) defines the point
−→
q H of the topos [0, ∞) o N× associated to the abstract ordered rank one group H := ∪(∏k1 n j )−1 Z.
j−1
/ V. Let m j := ∏1 n` and H := ∪ m−
Proof. Let F = lim y I j . One has F (V ) = ∅ when 0 ∈ 1
j Z.
−→
We denote by ι( j, k) := k/m j ∈ H the element of H given by the image of k ∈ Z by the canonical
injection Z → H associated to the j-th copy of Z in the colimit. One has by construction
ι( j, k) = ι( j + 1, n j k) , ∀ j, k ∈ Z. (7)
Let V be a bounded open interval with 0 ∈ V. One then obtains an injection
k
α : F (V ) → H+ , α ( I j → V ) := ι( j, k)
by using the compatibility with the inductive limits, i.e.
k nj
α ( I j → V ) ◦ ( I j+1 → I j ) = ι( j + 1, n j k) = ι( j, k).
Let us show that α is surjective. Given k ∈ N× and j > 0 we prove that for ` large enough one
has k ∏`−
j
1
ni ∈ y I` (V ). Let > 0 such that the neighborhood W of 0 of radius is contained
in V. Then for ` large enough, using the hypothesis (5), one gets that k ∏`−
j
1
ni I` ⊂ W (since it
contains 0 and is of small enough diameter) and hence k ∏`−
j
1
ni ∈ y I` (V ).
Proposition 3.13. Let H be an abstract rank one ordered group. Let n j ∈ N× , I j be bounded open
intervals as in Lemma 3.12 such that H := ∪(∏k1 n j )−1 Z. Then, the pullback part of the point q H of the
topos [0, ∞) o N× is given by the functor which associates to an object F of Sh(C , J ) the colimit (6).
The proof is the same as the one of Proposition 3.11.
16
4.1 The stalks of the structure sheaf O
The next result determines the structure of the stalks of O .
Theorem 4.2. (i ) At the point p H of the topos [0, ∞) o N× associated to the rank one subgroup H ⊂ R
the stalk of the structure sheaf O is the semiring R H of germs, at λ = 1, of Rmax -valued, piecewise affine
convex functions f (λ ) with slopes in H.
(ii) The stalk of the structure sheaf O at the point q H of [0, ∞) o N× associated to the abstract rank one
ordered group H is the semiring Z H associated by the max-plus construction to the totally ordered group
R × H and endowed with the lexicographic order.
Proof. (i ) To evaluate the stalk of the structure sheaf O at the point p H we use the description
given by (6). We let h j , n j , I j as in Proposition 3.11 and evaluate the colimit
n
OpH = lim O( Ik ), O( Ik+1 →k Ik ) : O( Ik ) → O( Ik+1 ).
−→
k
Thus the map ( j, f ) 7→ ρ( j, f ) is compatible with the colimit and determines a map ρ : OpH →
R H which is easily shown to be an isomorphism of semirings.
j
(ii) Let (n j ), with I j be as in Lemma 3.12 and such that H = ∪(∏1 n` )−1 Z. We let, as above, m j :=
j−1
∏1 n` so that H := ∪ m− 1
j Z. We set ι ( j, k ) : = k / m j ∈ H so that (7) follows by construction.
Then by Proposition 3.13, the stalk of the structure sheaf O at the point q H is the colimit OqH =
lim O( I j ). We define a map δ : OqH → H as follows. We associate to ( j, f ), with f ∈ O( I j ), the
−→
element δ ( j, f ) := ι( j, k) where k = f 0 (0) ∈ Z is the derivative of f at 0 ∈ I j . One then has
nj
δ ( j + 1, O( I j+1 → I j )( f )) = ι( j + 1, ( f (n j λ ))0λ =0 = ι( j + 1, n j f 0 (0)) = ι( j, f 0 (0)) = δ ( j, f ).
This shows that the map δ is well defined. Similarly, the equality α ( j, f ) := f (0) defines a map
α : OqH → Rmax and the pair ρ = (α, δ ) gives a map OqH → Z H which is both injective and
17
surjective. One easily checks that this map is an isomorphism of OqH for the semiring structure
whose multiplication corresponds to ( x, h) • ( x0 , h0 ) = ( x + x0 , h + h0 ) and the addition to
0
( x, h) if x > x
( x, h) ∨ ( x0 , h0 ) := ( x0 , h0 ) if x0 > x
( x, h ∨ h0 ) if x = x0 .
Next, we describe the semiring R H of germs at λ = 1 of Rmax valued piecewise affine (con-
tinuous) convex functions f (λ ) with slopes in H. The germ of f is determined by the triple
( x, h+ , h− ), with x ∈ R and h± ∈ H given by x = f (1), f (1 ± ) = x ± h±, for ≥ 0 small
enough. The triples ( x, h+ , h− ) obtained from elements of R H are characterized by the condition
h+ ≥ h− which corresponds to the convexity of the function f (1 ± ) = x ± h± for ≥ 0 small
enough. The only other element of the semiring R H corresponds to the germ of the constant
function −∞. This function plays the role of the “zero" element for the following algebraic rules
applied to the non-zero elements of the semiring. The “addition" ∨ is given by the max between
two germs and hence it is described by the formula
0
( x, h+ , h− ) if x > x
( x, h+ , h− ) ∨ ( x0 , h0+ , h0− ) := ( x0 , h0+ , h0− ) if x0 > x
( x, h+ ∨ h0+ , h− ∧ h0− ) if x = x0 .
The “product" of two germs is given by their sum and hence it is described by the formula
The conditions on the functions f ( H ) on subgroups H ⊂ R which define the structure sheaf are
expressed locally in terms of the scaling flow by the two derivatives D± where
f ((1 + ) H ) − f ( H )
D± ( f )( H ) := lim (9)
→0±
and can be written in the form
D± ( f )( H ) ∈ H , ∀ H, D+ ( f )( H ) ≥ D− ( f )( H ) , ∀ H. (10)
18
Theorem 4.3. The canonical projection from the set S (Rmax ) of the points of the scaling site S defined
+ to the points of the topos [0, ∞ ) o N is bijective.
over Rmax ×
f ∨ g = f ⇒ φ( f ) = φ( f ∨ g) = φ( f ) ∨ φ( g) ⇒ φ( f ) ≥ φ( g).
This shows that one cannot have φ( f ) = −∞. This argument shows that for any elements f , g
of R H , one has f < g ⇒ φ( f ) ≤ φ( g) and it follows that
( x − y) ∨ ( z − t) = (( x + t) ∨ ( y + z)) − ( y + t).
19
As in the case of R H the germ of a function f is characterized by the triple ( x, h+ , h− ) with x ∈ R
and h± ∈ H given by x = f (1), f (1 ± ) = x ± h±, for ≥ 0 small enough. The triples
( x, h+ , h− ) obtained from elements of Frac R H correspond to arbitrary values of ( x, h+ , h− ) ∈
R × H × H. As above the only other element of Frac R H corresponds to the germ of the con-
stant function −∞ which plays the role of the “zero" element. The algebraic rules for the other
elements are given for the “addition" ∨ of two germs by the max of the two germs and hence by
0
( x, h+ , h− ) if x > x
( x, h+ , h− ) ∨ ( x0 , h0+ , h0− ) := ( x0 , h0+ , h0− ) if x0 > x (11)
( x, h+ ∨ h0+ , h− ∧ h0− ) if x = x0
The “product" of the germs is given by the sum of the two germs and hence by
( a ∨ b) − (c ∧ d) ≥ ( a − c) ∨ (b − d) ≥ ( a − c) ∧ (b − d) , ∀ a, b, c, d ∈ R.
Note that one needs the ∧ in (13) to take care of the cases x 6= x0 , but when x = x0 the ∨ works.
Finally, the germ of f + g is given by the sum of the two germs and hence (14) holds.
20
Lemma 5.1. (i ) The map R∗+ → C p , λ 7→ λH p induces the isomorphism η p : R∗+ / pZ → C p .
(ii) The pull-back by η p of the (restriction to C p of the) structure sheaf O is the sheaf O p on R∗+ / pZ of
piecewise affine (continuous), convex functions with slopes in H p .
Proof. (i ) For λ ∈ R∗+ one has λH p ∈ C p . An abstract isomorphism of ordered groups φ : H p →
H ⊂ R is given by multiplication by φ(1) = λ. Since pH p = H p the map R∗+ → C p induces a
surjective map η p : R∗+ / pZ → C p . We show that η p is injective. If λH p = λ 0 H p , then µH p = H p
for µ = λ /λ 0 . Thus µ = a/ pn ∈ H p and the same result holds for µ −1 , thus µ is a power of p.
(ii) The condition on the functions f ( H ) on subgroups H ⊂ R which defines the structure sheaf
are expressed locally in terms of the scaling flow by the two derivatives D± of (9) and can be
written in the form (10). On the function g = f ◦ η p , g(λ ) := f (λH p ), condition (10) becomes
λ∂± + − ±
λ g ( λ ) ∈ λH p , λ∂λ g ( λ ) ≥ λ∂λ g ( λ ) (where the ∂λ are the directional derivatives) or equiva-
lently: ∂± + −
λ g ( λ ) ∈ H p and ∂λ g ( λ ) ≥ ∂λ g ( λ ) thus one gets (ii ).
5.1 Divisors
Proposition 5.2. (i ) The sheaf of quotients of the sheaf of semirings O p is the sheaf K p on R∗+ / pZ of
piecewise affine, continuous functions with slopes in H p , endowed with the two operations max, +.
(ii) The quotient sheaf of Cartier divisors C aC`(C p ) := K ×p /O p is isomorphic to the sheaf Div (C p ) of
×
∀λ ∈ R∗+ , ∃V open λ ∈ V : D (µ ) = 0 , ∀µ ∈ V, µ 6= λ.
21
λ� λ� λ�=�λ�
��� ��� �� ��
Figure 2: A function f ∈ K × (C p ).
∑ Ord H ( f ) = 0 , ∀ f ∈ K(C p ).
H
By applying the equality above when j = n, and using the periodicity property f (λn ) = f ( pλ0 ) =
f (λ0 ) one obtains
n
∑(λi − λi−1 )hi = 0.
1
22
By using λn = pλ0 one also has
n−1
( λ 1 − λ 0 ) h1 + ( λ2 − λ1 ) h2 + . . . + ( λ n − λ n−1 ) hn = ∑ λi (hi − hi+1 ) + λ0 ( phn − h1 ).
1
For 1 ≤ i ≤ n − 1 one has λi (hi − hi+1 ) = −Order( f )(λi ) where we set Order( f )(λ ) := OrdλHp f .
Moreover, notice that phn is the slope of the function f in the interval (1/ p) In so that the order of
f at λ0 is λ0 (h1 − phn ). Then, the above equality shows that the sum of all orders must vanish.
Next, we consider the problem of constructing a global section f ∈ K(C p ) whose divisor is an
assigned divisor of degree zero.
Let J p := ( p − 1) H p ⊂ H p be the principal ideal generated by the integer ( p − 1) in the ring H p .
One has the following exact sequence of rings
χ
0 → J p → H p → Z/( p − 1)Z → 0
23
Proof. We show first that any divisor D such that deg( D ) = 0 and χ( D ) = 0 is principal. Indeed,
the divisor D can be written in the form
where the elements a j ∈ H p are such that ∑n0 −1 a j λ j = 0. Thus one is given the intervals I j =
[λ j−1 , λ j ] with λn = pλ0 . In order to show that D is principal one needs to find elements h j ∈ H p
for 1 ≤ j ≤ n, giving the slope of f on I j and hence such that
The solution of the above system of equations is unique and of the form
This argument shows that one can extend f by periodicity so that f ( pλ ) = f (λ ) and obtain a
continuous, piecewise affine function with slopes in H p . Moreover, by construction the divisor
of f is D, since the discontinuities of the derivative take place at the λ j ’s and are given by the
a j ’s. It remains to show that the restriction of the map χ to the subgroup Div0 (C p ) ⊂ Div( D )
of divisors of degree 0 is surjective onto Z/ pZ. Given m ∈ Z/( p − 1)Z, it is enough to find an
increasing finite sequence of real numbers λ j > 0, λ0 < λ1 < . . . < λn = pλ0 and elements
a j ∈ H p such that ∑0n−1 a j λ j = 0 and ∑n0 −1 a j = m mod. ( p − 1) H p . Both the group H p ⊂ R and
(m)
the subset H p := { x ∈ H p | x = m mod. ( p − 1) H p } are dense in R. This shows that one can
fix arbitrarily the (λ j ) and all the a j except for one of them, say ak , and then choose ak ∈ H p with
preassigned χ( ak ) ∈ Z/( p − 1)Z such that for a choice of λk0 close to λk one gets
n−1 n−1
∑ a j λ0j = 0, ∑ χ(a j ) = m ∈ Z/( p − 1)Z.
0 0
The same argument also shows that one can arbitrarily prescribe both deg( D ) and χ( D ) for
divisors D ∈ Div(C p ). Since the group law on divisors is given by pointwise addition of sections,
both maps deg : Div(C p ) → R and χ : Div(C p ) → Z/( p − 1)Z are group homomorphisms and
one obtains the isomorphism of groups (16).
5.2 Symmetries
The scaling site S inherits from its construction by extension of scalars, as in Section 2.1, several
symmetries such as the arithmetic Frobenius associated to the automorphisms of Rmax + ∼ Rmax
24
or the absolute Frobenius associated to the frobenius endomorphisms of the semirings of charac-
teristic one of the structure sheaf. These symmetries induce symmetries of the curves C p and in
this subsection we describe them as operators acting on the structure sheaf O p of these curves.
Since the divisors on C p are Cartier divisors, one then obtains, as a byproduct, induced actions
on Div(C p ) and on its quotient Div(C p )/P .
Arithmetic Frobenius
The analogue of the arithmetic Frobenius Fr a on sections f of O p is defined as follows
This operator preserves the properties of f ( f is convex, piecewise affine with slopes in H p and
periodic f ( pλ ) = f (λ )) as well as the algebraic operations (∨, +). Thus it defines automor-
phisms denoted by Frµa . On a function f which is locally written as ∨(n j λ + a j ) the action of Frµa
replaces the a j ’s by µa j and this corresponds to the arithmetic Frobenius. The induced action on
the stalks of O p associates to a germ at λ a germ at µλ given by x 7→ µ f (µ −1 x), for x near µλ.
One thus obtains a morphism from germs at H to germs at µH
Indeed, one has f ((1 + ) H ) ∼ f ( H ) + h+ , for > 0 small and thus
Next, we list some properties of the induced action by Frµa on divisors. We use the notation D =
∑( H j , h j ), with h j ∈ H j for the divisor with support on the set { H j } and such that D ( H j ) = h j
∀ j.
Lemma 5.7. The action induced by Frµa on divisors is given by
25
Relative Frobenius
Let h ∈ H p , h > 0. One defines the operator Frrh acting on functions by
Frrh ( f )(λ ) := f (hλ ) , ∀λ , ∀h ∈ H p+ .
This operator acts on a section f of O p locally written as f = ∨(n j λ + a j ) by replacing the n j
by hn j , while leaving the a j unchanged. In particular it is Rmax -linear. Since the powers pn act
trivially in view of the periodicity of f , the operator Frrh only depends upon the class of h in the
quotient H p+ / pZ which is the multiplicative monoid N× ( p)
of positive integers relatively prime
to p. The induced action on the stalks associates to a germ at λ, the germ at h−1 λ given by
x 7→ f (hx) for x near h−1 λ. One thus obtains a morphism from germs at H to germs at h−1 H
Frrh : R H → Rh−1 H , ( x, h+ , h− ) 7→ ( x, h+ , h− ) ∈ Rmax × h−1 H × h−1 H (18)
as follows from the identity Frrh ( f )((1 + )h−1 H ) = f ((1 + ) H ).
Lemma 5.8. The action induced by Frrh on divisors is given by
Absolute Frobenius
This operator acts on sections f of O p by composition with the Frobenius of Rmax which, in this
logarithmic notation, multiplies the function f by a constant. In order to preserve the property
of the slopes in H p one takes this constant to be an element in H p . More precisely, the action of
the absolute Frobenius on functions is given, for any h ∈ H p+ , by the formula
Frh ( f )(λ ) := h f (λ ) , ∀λ , ∀h ∈ H p+ .
Its properties follow from the properties of the two previous operators since
Frh = Frha ◦ Frrh = Frrh ◦ Frha , ∀h ∈ H p+ . (19)
Indeed, one has Frha ◦ Frrh ( f )(λ ) = h f (h−1 hλ ) = h f (λ ) and similarly Frrh ◦ Frha = Frh .
Lemma 5.9. The action induced by Frh on divisors is given by
D= ∑( H j , h j ) 7→ Frh (D) = ∑( H j , hh j ).
This action is trivial on the points, i.e. it fixes Support( D ). It preserves the subgroup P of principal divi-
sors, acts on the invariant χ by multiplication by χ(h) ∈ Z/( p − 1)Z and on the degree by multiplication
by h ∈ H p+ ⊂ R∗+ .
26
��
Θ(λ)∼λ/(�-�)-��� λ/��� �
�
� �� �� �� ��
Proof. This follows from (19) using Lemmas 5.7 and 5.8.
Notice that each point H ∈ C p determines a morphism
and the absolute Frobenius is compatible with this morphism: p H (Frh ( f )) = Frh ( p H ( f )). In
particular it acts trivially on the spectrum.
n2 −n
∞
θ (w, t) = ∑(−1)n t 2 wn = (1 − w) ∏(1 − tm )(1 − tm w)(1 − tm w−1 ) (20)
Z 1
and satisfies the functional equation: −wθ (tw, t) = θ (w, t). Its relation with the standard theta
√
function ϑ1 (u, q) is given by the equation: ϑ1 (u, q) = i 4 q e−iuθ (e2iu , q2 ). This suggests to trans-
pose (20) naively in order to define theta functions for C p = R∗+ / pZ . In our framework, the
role of the product is replaced by addition, thus the following infinite sums replace the infinite
products on the left
∞ ∞
∏(1 − tm w) f + (λ ) := ∑ (0 ∨ (1 − pm λ))
0 0
∞ ∞
∏ ( 1 − tm w−1 ) 0 ∨ ( p−m λ − 1 ) .
f − (λ ) := ∑
1 1
27
Lemma 5.10. (i ) The functions f ± (λ ), λ ∈ (0, ∞) are convex (continuous), piecewise affine with slopes
in H p and
f + ( pλ ) − f + (λ ) = − (0 ∨ (1 − λ )) , f − ( pλ ) − f − (λ ) = (0 ∨ (λ − 1)) . (21)
(ii) The function θ (λ ) := f + (λ ) + f − (λ ), λ ∈ (0, ∞) (cf. Figure 3), is convex (continuous), piecewise
affine with slopes in H p and fulfills the equation: θ ( pλ ) = θ (λ ) + λ − 1, ∀λ ∈ (0, ∞).
(iii) One has
1
|θ (λ ) − λ − log λ / log p | ≤ 1 ∀λ ∈ (0, ∞).
p−1
Proof. (i ) Notice that the sum ∑∞ m
0 ( 0 ∨ ( 1 − p λ )) has only finitely many non-zero terms since
m
p λ > 1 for m large enough. Each of these terms is convex continuous, piecewise affine with
slopes in H p and thus the same property holds for f + . Moreover, the difference f + ( pλ ) − f + (λ )
is given by the single term corresponding to m = 0, i.e. f + ( pλ ) − f + (λ ) = − (0 ∨ (1 − λ )).
Similarly for f − (λ ) = ∑∞ 1 (0 ∨ ( p
−m λ − 1 )) one gets the term (0 ∨ (λ − 1 )) in f ( pλ ) − f (λ ).
− −
Thus we obtain (21).
(ii) The first part of the statement follows from (i). Moreover one has
θ ( pλ ) − θ (λ ) = f + ( pλ ) − f + (λ ) + f − ( pλ ) − f − (λ ) = − (0 ∨ (1 − λ )) + (0 ∨ (λ − 1)) = λ − 1
Θh,µ (λ ) := µ θ (µ −1 hλ ).
It is a convex continuous, piecewise affine function with slopes in H p and fulfills the equation
µ θ (µ −1 hpλ ) = µ θ (µ −1 hλ ) + µ µ −1 hλ − µ.
It follows that
Θ ph,µ (λ ) = Θh,µ (λ ) + hλ − µ.
To a pair of labels (h, µ ) we associate the divisor which is everywhere zero except on the sub-
group H = µh−1 H p (abstractly isomorphic to H p ) where it takes the value µ ∈ H. We let
δ (h, µ ) := (µh−1 H p , µ ).
28
Note that by construction one has δ ( ph, µ ) = δ (h, µ ), so that δ (h, µ ) only depends upon the class
of h modulo the multiplication by powers of p. Such a class is uniquely specified as that of an
integer m > 0, prime to p. Given a simple divisor, i.e. a positive divisor supported on a single
H, and µ ∈ H, µ > 0 one can find λ > 0 such that H = λH p and thus h ∈ H p with µ = hλ. One
then has δ (h, µ ) = ( H, µ ). The choice of λ (and hence of h) is unique only up-to multiplication
by a power of p.
The classical description of elliptic functions in terms of theta functions admits the following
counterpart in our framework, which provides, in particular, with another proof of Theorem 5.6
by an explicit construction of an f ∈ K(C p ) with given divisor.
Proposition 5.11. Let D = D+ − D− ∈ Div(C p ) be a divisor (D± ∈ Div+ ) and (hi , µi ) ∈ H p+ × R∗+ ,
(h0j , µ 0j ) ∈ H p+ × R∗+ such that D+ = ∑ δ (hi , µi ) and D− = ∑ δ (h0j , µ 0j ). Then, if deg( D ) = 0 and
h ∈ H p fulfills ( p − 1)h = ∑ hi − ∑ h0j , the following function
is continuous, piecewise affine with slopes in H p , fulfills f ( pλ ) = f (λ ) ∀λ ∈ R∗+ and one has: Div( f ) =
D.
Proof. By applying the equation (22), one has
Theorem 5.12. Let f ∈ K(C p ). Then f admits the following canonical decomposition
29
5.4 Riemann-Roch theorem of type II
In this subsection we define a Riemann-Roch problem for divisors on the curves C p and prove
a Riemann-Roch formula. To this end, we introduce the notion of continuous dimension for the
Rmax -modules H 0 (C p , O( D )) associated to divisors D on C p .
It follows from Proposition 5.2 that the notion of divisors on C p coincides with the notion of
global section of the sheaf C aC`(C p ) = K ∗p /O ∗p of Cartier divisors. In analogy with the classical
case, a Cartier divisor D described by local sections f i ∈ K ∗p (Wi )/O ∗p (Wi ), defines a subsheaf
O( D ) of K p . This is the sheaf of O p -modules generated on Wi by the (multiplicative) inverses
of the f i , i.e. in the additive notation of this tropical set-up by the functions − f i . The sections
of O( D ) are given locally by the rational functions f ∈ K p (W ) which satisfy the inequality
D + ( f ) ≥ 0, where ( f ) denotes the principal divisor associated to f . By construction, O( D ) is a
sheaf of O p -modules, and in particular its global sections define a module over Rmax
The slopes of the functions f ∈ K(C p ) can have arbitrarily small real size since the group H p ⊂ R
is dense. This shows that when deg( D ) > 0, (25) will in general yield an infinite dimensional
space of solutions. However, notice that the group H p is discrete when embedded diagonally in
Q p × R, and this fact allows one to obtain a natural norm on sections of K p by implementing the
p-adic norm of the slopes. In the following, we choose to normalize the p-adic norm |h| p ≥ 0,
∀h ∈ H p so that | p| p = 1/ p.
Let f be a continuous, piecewise affine function on R∗+ with slopes h± (u) ∈ H p and such that
f ( pu) = f (u). The slope h± ( pλ ) of f at pλ is h± (λ )/ p since
f ( pλ + ) − f ( pλ ) f ( pλ + pδ ) − f ( pλ ) 1 f (λ + δ ) − f (λ )
h± ( pλ ) := lim = lim = lim .
→0± δ →0± pδ p δ →0± δ
The value of |h± (λ )| p /λ is unchanged if one replaces λ by pλ since the p-adic norm of h± (λ )/ p
is p|h± (λ )| p .
30
Definition 5.14. Let f ∈ K(C p ). We set
Note that the value |h(1)− | p corresponding to the ingoing slope at λ = 1 is taken into account
for λ ∈ [1, p] near p as the limit of the values |h(λ )| p /λ when λ → p, since h( p)− = h(1)− / p.
If the restriction of f to [1, p] has integral slopes, one has |h(λ )| p ≤ 1 for all λ ∈ [1, p] and
thus k f k p ≤ 1. Conversely, if k f k p ≤ 1 one has |h(λ )| p < p for all λ ∈ [1, p) and thus, since
h(λ ) ∈ H p , one gets h(λ ) ∈ Z for all λ ∈ [1, p) as required.
(v) It follows from (i) and (ii) that H 0 ( D )ρ is an Rmax -submodule of H 0 ( D ), moreover one easily
0
sees that H 0 ( D )ρ ⊂ H 0 ( D )ρ for ρ < ρ0 .
The next step is to define the continuous dimension of the module H 0 ( D ) using the filtration by
the submodules H 0 ( D )ρ by means of a formula of the form
1
DimR ( H 0 ( D )) := lim dim( H 0 ( D )ρ ),
ρ→∞ ρ
where, on the right hand side, one uses a suitable notion of integer valued dimension for Rmax -
modules. In our context, the most natural notion is that of “topological dimension" that counts
the number of real parameters on which a general element depends. The original definition for
such dimension is due to Lebesgue. We use the reference [15].
5 at a point of discontinuity of the slopes one takes the max of the two values |h± (λ )| p /λ in (26)
31
Definition 5.16. Let X be a topological space. The topological dimension dimtop ( X ) of X is the small-
est integer n such that every open cover U of X admits a refinement V such that every point of X is in at
most n + 1 elements of V .
When working with the modules H 0 ( D ) we use the topology of uniform convergence for Rmax -
valued functions on the interval [1, p] (or equivalently by periodicity on R∗+ ). The distance defin-
ing this topology is given by
We define
n
DimR ( H 0 ( D )) := lim p−n dimtop ( H 0 ( D ) p ) (28)
n→∞
Our next goal is to prove that the above limit exists and one obtains a Riemann-Roch formula.
Theorem 5.17. (i ) Let D ∈ Div(C p ) be a divisor with deg( D ) ≥ 0. Then the limit in (28) converges
and one has DimR ( H 0 ( D )) = deg( D ).
(ii) The Riemann-Roch formula holds
The proof of Theorem 5.17 will be given later and follows by combining Lemma 5.18 below with
the understanding of H 0 ( D ) in the special case discussed in Lemma 5.19.
Lemma 5.18. Let D ∈ Div(C p ) and f ∈ K(C p ). Then
(i) For D0 = D + ( f ) and for any non-negative integer n such that k f k p ≤ pn , the map H 0 ( D ) →
n n
H 0 ( D 0 ), ξ 7→ ξ − f induces an isomorphism of H 0 ( D ) p with H 0 ( D 0 ) p .
(ii) The absolute Frobenius f 7→ pn f determines a twisted isomorphism of Rmax -modules
n
Fpn : H 0 ( D ) p → H 0 ( pn D )1 .
(ii) The twisting is by the Frobenius Fr pn ∈ Aut(Rmax ) and occurs since Fpn ( f + a) = Fpn ( f ) +
pn a for any scalar a ∈ Rmax . This operation does not affect the topological dimension. The
principal divisor ( pn f ) is equal to pn × ( f ) and one has, using Proposition 5.15 (iii )
D + ( f ) ≥ 0 ⇐⇒ pn D + ( pn f ) ≥ 0, k f k p ≤ pn ⇐⇒ k pn f k p ≤ 1.
32
�(����� �������)
Lemma 5.19. (i ) The module E N,p is the Rmax -module of convex (continuous), piecewise affine functions
on [1, p] with integral slopes, such that f (1) = f ( p) and − f +0 (1) + p f −0 ( p) ≤ N.
(ii) Let a ∈ {1, . . . , N − p} and denote by b = E(( N − a)/ p) ≥ 1 the integer part of ( N − a)/ p. Let
φ a ( x) := max{− a( x − 1), b( x − p)}, for x ∈ [1, p]. Then φ a ∈ E N,p .
(iii) For > 0, let ∆N − p := {(t1 , . . . , t N − p ) | 0 < t1 < . . . < t N − p < }. The following map is
continuous and injective for sufficiently small (φ0 := 0 by convention)
j
N− p
h: R × ∆N − p → E N,p , h(t0 , . . . , t N − p ) := ∨0 (φ N − p− j − ∑ ti ) .
0
(ii) By construction, the function f = φ a is continuous, convex, piecewise affine with integral
a+bp
slopes and φ a (1) = φ a ( p) = 0. Moreover, at the point x = a+b where − a( x − 1) = b( x − p),
the slope of φ a changes from − a to b. The point x is inside the interval (1, p) since a > 0, b > 0.
Thus the slope of φ a is − a near 1 and b near p. The condition − f +0 (1) + p f −0 ( p) ≤ N is fulfilled
since a + pb ≤ N.
p−1
(iii) Let ≤ N − p+1 . Then on [ 1, 1 + ] all the φ a ( x )’s coincide with − a ( x − 1 ) since with b =
E(( N − a)/ p) ≥ 1 and a ∈ {1, . . . , N − p} one has
a + bp b( p − 1) p−1
−1 = = ( p − 1)(1 + a/b)−1 ≥ .
a+b a+b N−p+1
33
j N− p
For 0 ≤ j < N − p, let ψ j := φ N − p− j − ∑0 ti and let ψ N − p = − ∑0 ti be a constant function.
For 0 ≤ j ≤ N − p one has
j
ψ j ( x) = ( j + p − N )( x − 1) − ∑ ti , ∀ x ∈ [1, 1 + ].
0
where lim and lim denote respectively the lim inf and the lim sup. Let α j ∈ H p ( j = 1, 2),
α j > 0, be such that δ − ≤ α1 < δ, δ < α2 ≤ δ + and that χ(α j ) = χ( D ). Then let (as in
34
Lemma 5.13, (ii )) Pj ∈ Div+ (C p ) be positive divisors such that χ( Pj ) = 0 with deg( P1 ) = δ − α1 ,
deg( P2 ) = α2 − δ. One has deg(α1 {1} + P1 ) = δ and χ(α1 {1} + P1 ) = χ( D ). Thus by applying
Theorem 5.6 we get the existence of a function f 1 ∈ K(C p ) such that α1 {1} + P1 = D + ( f 1 )
and a function f 2 ∈ K(C p ) such that D + ( f 2 ) + P2 = α2 {1}. Using the natural injective maps,
isometric for the distance (27), given by the inclusions
+ f1
H 0 (α1 {1}) ⊂ H 0 (α1 {1} + P1 ) = H 0 ( D + ( f 1 )) → H 0 ( D )
as soon as k f 2 k p ≤ pn . Thus the result will follow provided we show that for any α ∈ H p , α > 0
one has
n
lim p−n dimtop ( H 0 (α {1}) p ) = α.
n→∞
n
By Lemma 5.18 (ii ), one has dimtop ( H 0 (α {1}) p ) = dimtop ( H 0 (α pn {1})1 ) and for n large
enough so that α pn is an integer, Lemma 5.19 (iv) shows that dimtop ( H 0 (α pn {1})1 ) = α pn −
p + 1. Thus one gets as required
n
lim p−n dimtop ( H 0 (α {1}) p ) = lim p−n (α pn − p + 1) = α.
n→∞ n→∞
(ii) Follows from (i) and the fact that H 0 ( D ) = 0 if deg( D ) < 0.
Appendices
A The structure of the Rmax -module E N,p
In this appendix we provide further properties of the Rmax -module E N,p of convex, piecewise
affine, continuous functions f on [1, p] with integral slopes such that f (1) = f ( p) and − f 0 (1) +
p f 0 ( p) ≤ N. The convexity of f implies that its graph is above its tangent at any point i.e. one
has the inequality
f (u) ≥ f (v) + f 0 (v)(u − v) , ∀u, v ∈ [1, p]. (29)
If v is a point of discontinuity of the derivative, one can take for f 0 (v) any value between the left
and right limits f −0 (v) ≤ f +0 (v).
We recall the definition of the extremal rays of an Rmax -module.
Definition A.1. Let E be an Rmax -module, and f ∈ E . Then f is extremal iff the equality f 1 ∨ f 2 = f
implies that one of the f j = f . The Rmax -submodule { f + x | x ∈ Rmax } is called an extremal ray.
35
��� ��� ��� ���
-���
-���
ϕ�
-���
ϕ�
ϕ�
-���
-���
The following proposition describes the structure of the Rmax -module E N,p . We use the notation
of Lemma 5.19 (ii).
Proposition A.2. The functions φ a , for 0 ≤ a ≤ N − p, are extremal elements of E N,p and the elements
of the set {φ a + x | x ∈ Rmax } are the extremal rays of E N,p . These elements generate the module E N,p ,
i.e. the following map is surjective
N − p+1
σ : Rmax → E N,p , x = ( x a ) 7→ σ ( x) := ∨(φ a + x a ). (30)
Proof. We use Lemma A.3 proved below. We prove first that φ a is extremal. Assume that φ a =
f 1 ∨ f 2 , with f j ∈ E N,p for j = 1, 2. Then one of the f j , say f 1 , agrees with φ a in a neighborhood
V 3 λ = 1. By (33) one has f 1 = ∨(φ` + t` ) for some real numbers t` . One of the φ` + t` agrees
with f 1 in a neighborhood of 1. Comparing the slopes one gets ` = a and t a = 0. Thus one
derives that f 1 ≥ φ a and since φ a = f 1 ∨ f 2 ≥ f 1 one concludes that f 1 = φ a . This argument
shows that φ a is extremal. The equality (33) shows that any extremal element of E N,p is of the
form φ a + t for some a and t ∈ R so that the φ a generate all the extremal rays. Finally (33) shows
that σ is surjective.
For a ∈ {0, . . . , N − p}, let γ a : E N,p → Rmax be defined by
Lemma A.3. (i ) Let f ∈ E N,p and x ∈ (1, p). Then, there exists a ∈ {0, . . . , N − p} such that
36
Proof. (i ) If f is constant (32) holds for a = 0, thus we assume that f is non-constant. Let α =
− f 0 (1) and apply (29) for v = 1, u = p. Using the equality f (1) = f ( p) one has α ≥ 0, and in
fact α > 0 since f is non-constant. Let β = f 0 ( p), by applying (29) for v = p, u = 1 and using
f (1) = f ( p) one has β ≥ 0, and in fact β > 0 since f is non-constant. Furthermore one has
α + pβ ≤ N since f ∈ E N,p . By continuity of the functions involved and the finiteness of the
set {0, . . . , N − p}, one can assume that f is affine at x. Let us first assume that f 0 ( x) ≤ 0. Let
a = − f 0 ( x) so a ≥ 0. Since f is convex its derivative is non-decreasing and one has f 0 ( x) ≥ f 0 (1)
so that a ≤ α. Moreover by applying (29) for v = x, u = 1 one has
Thus
f ( x) ≥ f (1) + β( x − p) ≥ f ( x) + a( x − 1) + β( x − p)
and one gets a( x − 1) + β( x − p) ≤ 0, i.e. a( x − 1) ≤ β( p − x). One has α + pβ ≤ N and a ≤ α
so that a + pβ ≤ N and β ≤ ( N − a)/ p. But β = f 0 ( p) ∈ N is an integer ≤ ( N − a)/ p and
thus it is ≤ E(( N − a)/ p) = b. Then β ≤ b and one has a( x − 1) ≤ β( p − x) ≤ b( p − x). By
definition φ a ( x) := max{− a( x − 1), b( x − p)} and thus, since − a( x − 1) ≥ b( x − p), one gets
φ a ( x) = − a( x − 1). By applying (29) for v = x, u = z one has
f ( z) ≥ f ( x) + f 0 ( x)( z − x) = f ( x) − a( z − x) =
= f ( x) − a( z − 1) + a( x − 1) = f ( x) − a( z − 1) − φ a ( x).
Thus
f ( z) − f ( x) ≥ − a( z − 1) − φ a ( x) , ∀ z ∈ [1, p]. (34)
By applying (29) for v = p, u = z, one has
f ( z) ≥ f ( x) − φ a ( x) + b( z − p) , ∀ z ∈ [1, p] (35)
37
Since α + pβ ≤ N, one gets a ≥ α. Moreover one has f 0 ( x) > 0 and thus a ≤ N − p. The integer
b := E(( N − a)/ p) involved in the definition of φ a is equal to f 0 ( x) since N − a = p f 0 ( x) by
construction. Thus b = f 0 ( x). By applying (29) for v = 1, u = x one has
f ( p) ≥ f ( x) + f 0 ( x)( p − x) = f ( x) + b( p − x).
f ( x) ≥ f (1) − α ( x − 1) ≥ f ( x) + b( p − x) − α ( x − 1)
f ( z) − f ( x) ≥ f 0 ( x)( z − x) = b( z − x) = b( z − p) − b( x − p) = b( z − p) − φ a ( x).
Thus
f ( z) − f ( x) ≥ b( z − p) − φ a ( x) , ∀ z ∈ [1, p]. (36)
By applying (29) for v = 1, u = z one has, for z ∈ [1, p]
and since α ≤ a one gets −α ( z − 1) ≥ − a( z − 1), so that f ( z) ≥ f (1) − a( z − 1). By (36) for
z = p one has f ( p) − f ( x) ≥ −φ a ( x) and since f (1) = f ( p) one gets f (1) ≥ f ( x) − φ a ( x). It
follows that for z ∈ [1, p] one has
f ( z) ≥ f (1) − a( z − 1) ≥ f ( x) − φ a ( x) − a( z − 1)
and hence
f ( z) ≥ f ( x) − φ a ( x) − a( z − 1) , ∀ z ∈ [1, p]. (37)
Thus combining (36) and (37), one gets
38
Indeed, by (33) one has σ ◦ γ = Id, and since γ a ( f ) is the largest value of t for which φ a + t ≤ f
For f , g ∈ E N,p one has σ −1 ( f ∨ g) ⊃ σ −1 ( f ) ∨ σ −1 ( g). However the equality does not hold in
general which explains why γ does not preserve ∨ in general. Given f ∈ E N,p the fiber σ −1 ( f )
is of the form, with S f ⊂ {0, . . . , N − p} a suitable subset depending on f
N − p+1
σ −1 ( f ) = { x = ( x a ) ∈ Rmax | xa = γa ( f ) , ∀a ∈ S f , xa ≤ γa ( f ) , ∀a ∈
/ S f }.
The object M q N ' M × N is then called the biproduct of M and N, and denoted by M ⊕
N. Moreover matrix notation works for morphisms. In particular, the matrix Id M Id M
gives rise to a canonical element s M ∈ HomC ( M × M, M). For any pair of objects A, M of C ,
this construction turns HomC ( A, M) into a commutative monoid with a zero element. Given
α, β ∈ HomC ( A, M), their “sum" is obtained as the composition s M ◦ (α, β), where the pair
(α, β) is viewed as an element of HomC ( A, M × M) = HomC ( A, M) × HomC ( A, M). Moreover,
composition of morphisms is distributive over this addition and thus one has
Lemma B.2. Let C be a semiadditive category. Then for any object M of C the set EndC ( M) :=
HomC ( M, M) is canonically endowed with the structure of semiring.
39
Proposition B.3. (i ) If a semifield contains a non-trivial root of unity it is a field.
(ii) The only finite semifield which is not a field is B.
(iii) The only semifield whose multiplicative group is infinite cyclic is Zmax .
Proof. (i ) By Proposition 4.41 of [10], any element of finite order in a semifield F which is not a
field, is equal to 1. Indeed, with un = 1 one finds that a := 1 + u + . . . + un−1 fulfills ua = a so
that u = 1 if a 6= 0. If a = 0 one gets an element b := u + . . . + un−1 such that 1 + b = 0 and one
has a field.
(ii) By (i) all elements of a finite semifield F which is not a field are 0 or 1. Moreover the equation
1 + 1 = 1 holds thus F = B is the only finite semifield which is not a field.
(iii) Let K be a semifield whose multiplicative group is infinite cyclic. The map χ : N → K
defined inductively by χ(n + 1) = χ(n) + 1 is a morphism of semirings. Assume first that χ is
injective. Then χ(n) ∈ G for all n 6= 0, where G is the infinite cyclic multiplicative group of K.
0
Thus there would exist two distinct primes p, p0 and integers a, a0 > 0 such that χ( p) a = χ( p0 ) a
but this contradicts the injectivity of χ. Thus χ is not injective and we let n < m be such that
χ(n) = χ(m). Let k ∈ N and assume k ≥ m. Write k = am + b with a > 0, 0 ≤ b < m. Then one
has
χ(k) = χ( am + b) = χ( a)χ(m) + χ(b) = χ( a)χ(n) + χ(b) = χ( an + b) = χ(k0 )
where k0 = an + b < am + b = k. This shows that the image χ(N) is the same as the image
χ({0, . . . , m − 1} and is hence finite. Since the multiplicative group of K is infinite cyclic, it
follows that every element of χ(N) is in {0, 1} ⊂ K. There are two cases. If 1 + 1 = 0, then one
is in a field of characteristic 2 and taking a generator q of the multiplicative group one has an
equation of the form 1 + q = q a for some a ∈ / {0, 1} showing that K is an algebraic extension of
F2 and thus it is finite. If 1 + 1 = 1 one is in characteristic 1 and since K is a semifield the maps
Frn , Frn ( x) := xn , are injective endomorphisms ( [10] Proposition 4.43). Let then q be a generator
of the multiplicative group and a ∈ Z such that 1 + q = q a . One then has, using the maps Frn ,
2
that 1 + qn = qna for all n > 0. For a > 0 one has 1 + q a = q a and the associativity implies
2
q a = 1 + q a = 1 + (1 + q) = (1 + 1) + q = 1 + q = q a
40
slopes, gives a framework, common to all these cases, for the localization of zeros of analytic
functions in the punctured unit disk. The additional structure involved in the scaling site, i.e.
the action of N× by multiplication on the tropical half-line corresponds, as shown below in (39)
and (40), to the transformation on functions given by the composition with the n-th power of the
variable. The tropical notion of “zeros" of a convex piecewise affine function f with integral slope
is that a zero of order k occurs at a point of discontinuity of the derivative f 0 with k equal to the
sum of the outgoing slopes. The conceptual meaning of this notion is understood either by using
Cartier divisors or by implementing the natural hyperstructure on Rmax as shown in [19, 20].
Up to a trivial change of variables, this notion is well known in p-adic analysis where the function
−τ (− x) or rather its graph is called the valuation polygon of the series (cf. [16] chapter 6, §1.6).
This polygon is dual to the Newton polygon of the series which is, by definition, the lower
part of the convex hull of the points of the real plane with coordinates ( j, v( a j )), where v is the
valuation. By construction, τ ( f )( x) is finite since using the convergence hypothesis the terms
−nx − v( an ) tend to −∞ when |n| → ∞. Thus one gets a convex and piecewise affine function.
Moreover one has the multiplicativity property (cf. [16] chapter 6, Proposition 1.4.2)
τ ( f g)( x) = τ ( f )( x) + τ ( g)( x) , ∀ x ∈ (0, ∞),
and the classical result
Theorem C.2. Let f ( X ) = ∑ an X n be a Laurent series with coefficients in K, convergent in an annulus
A(r1 , r2 ) = { z ∈ K | r1 < | z| < r2 }. Then the valuations v( zi ) of its zeros zi ∈ A(r1 , r2 ) (counted with
multiplicities) are the zeros (in the tropical sense and counted with multiplicities) of the tropicalization
τ ( f ) in (− log r2 , − log r1 ).
This classical result is, for instance, deduced from the case of Laurent polynomials (see [6],
Lemma 1.6) using the approximation argument of [16], proof of Theorem 6.2.2.1. In particu-
lar one can take r1 = 0, r2 = 1 so that A(r1 , r2 ) is the punctured unit disk D (0, 1) \ {0}. In that
case the τ ( f ) are convex piecewise affine functions on (0, ∞) and one has the compatibility with
the action of N× on functions by f ( X ) 7→ f ( X n )
τ ( f ( X n ))( x) = τ ( f )(nx) , ∀ x ∈ (0, ∞), n ∈ N× . (39)
41
C.2 Jensen’s Formula
For complex numbers, unlike in the non-archimedean case, it is no longer true that for a generic
radius r, the modulus | f ( z)| is constant on the sphere of radius r and one replaces (38) by
Definition C.3. Let f ( z) be a holomorphic function in an annulus A(r1 , r2 ) = { z ∈ C | r1 < | z| < r2 }.
Its tropicalization is the function on the interval (− log r2 , − log r1 )
Z 2π
1
τ ( f )( x) := log | f (e−x+iθ )|dθ.
2π 0
For x ∈ (− log r2 , − log r1 ) such that f has no zero on the circle of radius e−x , the derivative
of τ ( f )( x) is the opposite of the winding number n( x) of the loop θ 7→ f (e−x+iθ ) ∈ C× . Thus
the function τ ( f )( x) is piecewise affine with integral slopes. When the radius e−x of the circle
increases the winding number of the associated loop increases by the number of zeros of f in the
intermediate annulus and this shows that the function τ ( f )( x) is convex and fulfills the Jensen’s
formula (cf. [17] Theorem 15.15) which is seen as the analogue of Theorem C.2.
Theorem C.4. Let f ( z) be a holomorphic function in an annulus A(r1 , r2 ) = { z ∈ C | r1 < | z| < r2 }
and zi ∈ A(r1 , r2 ) its zeros counted with their multiplicities. Then the values − log | zi | are the zeros (in
the tropical sense and counted with multiplicities) of the tropicalization τ ( f ) in (− log r2 , − log r1 ).
In particular one can take r1 = 0, r2 = 1 so that A(r1 , r2 ) is the punctured unit disk D (0, 1) \ {0}.
In that case the τ ( f ) are convex piecewise affine functions on (0, ∞) and one has the following
compatibility with the action of N× on functions by f ( z) 7→ f ( zn )
References
[1] M. Baker, S. Norine, Riemann-Roch and Abel-Jacobi theory on a finite graph, Advances in Math-
ematics 215 (2007), 766–788.
[2] A. Connes, C. Consani, The Arithmetic Site, Comptes Rendus Mathématiques Ser. I 352
(2014), 971–975.
[3] A. Connes, C. Consani, Geometry of the Arithmetic Site. Adv. Math. 291 (2016), 274–329.
[4] A. Connes, C. Consani, The scaling site. C. R. Math. Acad. Sci. Paris 354 (2016), no. 1, 1–6.
42
[5] J. Dixmier, On some C ∗ -algebras considered by Glimm. J. Functional Analysis, 1, (1967), 182–
203.
[6] B. Dwork, On the zeta function of a hypersurface. Inst. Hautes Etudes Sci. Publ. Math. No. 12
(1962) 5–68.
[7] M. Einsiedler, M. Kapranov, D. Lind, Non-Archimedean amoebas and tropical varieties. J. Reine
Angew. Math. 601 (2006), 139âĂŞ157.
[8] A. Gathmann, M. Kerber, A Riemann-Roch theorem in tropical geometry, Mathematische
Zeitschrift 259 (2008), 217-230
[9] S. Gaubert, Two lectures on Max Plus algebra. Proceedings of the 26th Spring School of Theo-
retical Computer Science, (1998), 83–147.
[10] J. Golan, Semi-rings and their applications, Updated and expanded version of The theory of
semi-rings, with applications to mathematics and theoretical computer science [Longman
Sci. Tech., Harlow, 1992. Kluwer Academic Publishers, Dordrecht, 1999.
[11] R. Hartshorne, Algebraic Geometry, Graduate Texts in Mathematics 52, Springer-Verlag, New
York Heidelberg Berlin 1977.
[12] S. Mac Lane, Categories for the working mathematician. Second edition. Graduate Texts in
Mathematics, 5. Springer-Verlag, New York, 1998.
[13] S. Mac Lane, I Moerdijk, Sheaves in geometry and logic. A first introduction to topos theory.
Corrected reprint of the 1992 edition. Universitext. Springer-Verlag, New York, 1994.
[14] G. Mikhalkin and I. Zharkov. Tropical curves, their Jacobians and theta functions. In Curves and
abelian varieties, volume 465 of Contemp. Math., p 203–230. Amer. Math. Soc., Providence,
RI, 2008.
[15] A. Pears Dimension theory of general spaces. Cambridge University Press, Cambridge,
England-New York-Melbourne, 1975.
[16] A. Robert, A course in p-adic analysis. Graduate Texts in Mathematics, 198. Springer-Verlag,
New York, 2000.
[17] W. Rudin, Real and complex analysis MacGraw-Hill, NewYork 1987.
[18] J. Tate, A review of non-archimedean elliptic functions. Elliptic curves, modular forms, and
Fermat’s last theorem (Hong Kong, 1993), 162–184, Ser. Number Theory, I, Int. Press, Cam-
bridge, MA, 1995.
[19] O. Viro, Hyperfields for Tropical Geometry I, Hyperfields and dequantization. Preprint. Arxiv:
math.AG/1006.3034, 2010.
[20] O. Viro, On basic concepts of tropical geometry. Tr. Mat. Inst. Steklova, 273, (Sovremennye
Problemy Matematiki): 271–303, 2011.
43